Model Fitting and Evaluation

Import Dataset

# import dataset
data_clean <- read_csv("data/data-clean.csv")

# quick check
head(data_clean, 10)

Cross-Validation Scheme

# set seed
set.seed(100)

# create initial split
splitted <- initial_split(data_clean, prop = 0.8, strata = "sentiment")

# quick check
splitted

#> <11681/2919/14600>

Defining Preprocess Recipe

# define preprocess recipe from train dataset
rec <- recipe(sentiment ~ ., data = training(splitted)) %>% 
  step_rm(-sentiment, -tweet) %>%
  step_string2factor(sentiment, levels = c("negative", "neutral", "positive"), skip = TRUE) %>%
  step_downsample(sentiment, ratio = 1/1, seed = 100) %>%
  step_mutate(tweet = str_squish(tweet)) %>% 
  step_mutate(tweet = replace_html(tweet, symbol = FALSE)) %>% 
  step_mutate(tweet = replace_kern(tweet)) %>% 
  step_mutate(tweet = replace_word_elongation(tweet)) %>% 
  step_mutate(tweet = replace_date(tweet, replacement = "datewords")) %>% 
  step_mutate(tweet = replace_time(tweet, replacement = "timewords")) %>% 
  step_mutate(tweet = replace_money(tweet, replacement = "moneywords")) %>% 
  step_mutate(tweet = replace_ordinal(tweet, remove = FALSE)) %>% 
  step_mutate(tweet = replace_number(tweet, remove = FALSE)) %>% 
  step_mutate(tweet = replace_internet_slang(tweet)) %>% 
  step_mutate(tweet = replace_contraction(tweet)) %>% 
  step_mutate(tweet = replace_emoji(tweet)) %>% 
  step_mutate(tweet = replace_symbol(tweet)) %>% 
  step_mutate(tweet = str_squish(tweet)) %>% 
  step_mutate(tweet = str_replace_all(tweet, "(<.*>)", "")) %>% 
  step_mutate(tweet = str_replace_all(tweet, "[:digit:]", "")) %>% 
  step_tokenize(tweet, token = "words") %>%
  step_stem(tweet) %>%
  step_stopwords(tweet) %>%
  step_tokenfilter(tweet, max_tokens = 256) %>%
  step_tfidf(tweet) %>%
  prep(string_as_factor = FALSE)

# get train and test dataset
data_train <- juice(rec)
data_test <- bake(rec, testing(splitted))

# quick check
head(data_train, 10)

Defining Model Specifications

# define model specification
model_spec <- boost_tree(
  mtry = 32,
  trees = 500,
  min_n = 1,
  tree_depth = 8,
  loss_reduction = 0.1,
  learn_rate = 0.1,
  sample_size = 0.8,
  mode = "classification"
)

# define model engine
model_spec <- set_engine(
  object = model_spec,
  engine = "xgboost",
  nthread = parallel::detectCores() / 2
)

# quick check
model_spec

#> Boosted Tree Model Specification (classification)
#> 
#> Main Arguments:
#>   mtry = 32
#>   trees = 500
#>   min_n = 1
#>   tree_depth = 8
#>   learn_rate = 0.1
#>   loss_reduction = 0.1
#>   sample_size = 0.8
#> 
#> Engine-Specific Arguments:
#>   nthread = parallel::detectCores()/2
#> 
#> Computational engine: xgboost

Model Fitting

# set seed
set.seed(100)

# fit the model
model <- fit_xy(
  object = model_spec,
  x = select(data_train, -sentiment),
  y = select(data_train, sentiment)
)

# quick check
model

#> parsnip model object
#> 
#> ##### xgb.Booster
#> raw: 2.7 Mb 
#> call:
#>   xgboost::xgb.train(params = list(eta = 0.1, max_depth = 8, gamma = 0.1, 
#>     colsample_bytree = 0.125, min_child_weight = 1, subsample = 0.8), 
#>     data = x, nrounds = 500, verbose = 0, objective = "multi:softprob", 
#>     num_class = 3L, nthread = 6)
#> params (as set within xgb.train):
#>   eta = "0.1", max_depth = "8", gamma = "0.1", colsample_bytree = "0.125", min_child_weight = "1", subsample = "0.8", objective = "multi:softprob", num_class = "3", nthread = "6", silent = "1"
#> xgb.attributes:
#>   niter
#> # of features: 256 
#> niter: 500
#> nfeatures : 256

Variable Importance

# get variable importance
var_imp <- xgb.importance(names(juice(rec, -sentiment)), model$fit)

# tidying
var_imp <- var_imp %>%
  arrange(desc(Gain)) %>% 
  head(10) %>% 
  rename(variable = Feature, importance = Gain) %>%
  mutate(variable = reorder(variable, importance))

# variable importance plot
ggplot(var_imp, aes(x = variable, y = importance)) +
  geom_col(fill = "darkblue") +
  coord_flip() +
  labs(title = "Variables Importance (Top 10)", x = NULL, y = NULL, fill = NULL) +
  scale_y_continuous(expand = expand_scale(mult = c(0, 0.1))) +
  theme_minimal()

Predict on Test Dataset

# predict on test
pred_test <- select(data_test, sentiment) %>%
  bind_cols(predict(model, select(data_test, -sentiment))) %>%
  bind_cols(predict(model, select(data_test, -sentiment), type = "prob"))

# quick check
head(pred_test, 10)

Confusion Matrix

# confusion matrix
pred_test %>%
  conf_mat(sentiment, .pred_class) %>%
  autoplot(type = "heatmap")

# metrics summary
pred_test %>%
  summarise(
    accuracy = accuracy_vec(sentiment, .pred_class),
    sensitivity = sens_vec(sentiment, .pred_class),
    specificity = spec_vec(sentiment, .pred_class),
    precision = precision_vec(sentiment, .pred_class)
  )

ROC Curve

# plot roc curve
pred_test %>%
  roc_curve(sentiment, .pred_negative:.pred_positive) %>%
  autoplot()

# get roc curve data on test dataset
pred_test_roc <- pred_test %>%
  roc_curve(sentiment, .pred_negative:.pred_positive)

# tidying
pred_test_roc <- pred_test_roc %>% 
  mutate_if(~ is.numeric(.), ~ round(., 4)) %>% 
  gather(metric, value, -.threshold, -.level)

# plot sensitivity-specificity trade-off
p <- ggplot(pred_test_roc, aes(x = .threshold, y = value)) +
  geom_line(aes(colour = metric)) +
  facet_wrap(~ .level, ncol = 1) +
  labs(x = "Probability Threshold to be Classified as Positive", y = "Value", colour = "Metrics") +
  theme_minimal()

ggplotly(p)

Precision-Recall Curve

# plot pr curve
pred_test %>%
  pr_curve(sentiment, .pred_negative:.pred_positive) %>%
  autoplot()

# get pr curve data on test dataset
pred_test_pr <- pred_test %>%
  pr_curve(sentiment, .pred_negative:.pred_positive)

# tidying
pred_test_pr <- pred_test_pr %>% 
  mutate_if(~ is.numeric(.), ~ round(., 4)) %>% 
  gather(metric, value, -.threshold, -.level)

# plot recall-precision trade-off
p <- ggplot(pred_test_pr, aes(x = .threshold, y = value)) +
  geom_line(aes(colour = metric)) +
  facet_wrap(~ .level, ncol = 1) +
  labs(x = "Probability Threshold to be Classified as Positive", y = "Value", colour = "Metrics") +
  theme_minimal()

ggplotly(p)